Record-only shadow mode
The first field use of Live Guidance is record-only: it logs every cue it would have spoken, without speaking or steering a single step, so its judgment can be audited against reality before it ever guides anyone.
iOS Vision — built and test-locked, in development builds today
It sees and describes
— spoken to you, in real time.
ChatLemur’s iOS Vision app turns the iPhone camera into a narrator: scenes, text, objects, colours and faces, described aloud — including Live Scene, continuous narration that speaks up only when the scene changes, and Point & Speak, which reads back just the text under your finger. The perception engine is built and locked by more than 2,300 automated tests, running in our internal development builds — on-device validation is the next step. Careful, staged navigation assistance is in development to follow.
In development The ambition: an all-in-one live navigation companion for blind people — bear-left/right, drop-off and edge detection, street crossings, path-finding, in one voice. To our knowledge, no one yet ships this on the phone a blind person already owns — including us; that’s what’s in development. It assists everything; it replaces nothing.
- Built — sees & describes
- In development — navigation
- Future — smart glasses
Development build — not yet available to download. It’s an assistive aid that supports your judgment; it never replaces your cane, your guide dog, or you.
Vision Tools — twelve accessibility tools (eight shown here), including “Who’s Here” (named faces).
The perception layer — built and test-locked
Built Point the phone at the world and it speaks. Twelve accessibility tools, each one a spoken answer to a real everyday question — built and test-locked, in a development build.
Twelve tools, one question each
Every tool answers out loud, in real time. The grid includes:
- Scene — what’s in front of me? (now also notes when a nearby seat looks free or occupied)
- Live Scene — keep narrating as the scene changes, not just once
- Point & Speak — point at a sign or label, hear just that text
- Find a Door — a hedged, camera-only door finder; never guesses distance or whether it’s open
- Read Document — read this text to me
- Money — which note is this?
- Find Object — where is the thing I need?
- Who’s Here — named faces, spoken
- Colours & objects — described aloud
- Read Text — any text in view, read aloud
- Light — is a light on or off, bright or dim?
- Barcode — scan a barcode, hear what it is
What the perception engine sees — boxes, labels, distances drawn on the scene.
Live Guidance — navigation that assists, never replaces
In development Spoken route cues with careful awareness of the path right in front of you — steps, kerbs, obstacles. This layer is in development, to assist route-following after supervised validation. It is not available yet.
Why this is the hard half. Apple — with LiDAR, the Neural Engine and Apple Intelligence in-house — ships the on-demand half: point, ask, hear a description. To our knowledge, nobody ships continuous, hands-free walking guidance on the phone a blind person already owns. We’ve built the on-demand half on our development iPhones; this layer is our attempt at the half nobody ships — in development, validated in stages with O&M specialists, and it will never say “it’s safe to cross.”
Live Guidance — the real screen, with the “Assistive aid only” note built into the app itself.
Start it, then walk
When it’s ready, you’ll start it and walk: Lenny is designed to speak the path, obstacles and distances aloud, with an annotated-display mode to show what it sees and a power-save mode that goes audio-only for long walks. In development — it’s meant to work by voice as you walk, not a screen to watch; cues heard here are simulated.
Hear the guidance system.
Development build — simulated cues. It assists your judgment; it never replaces your cane, your guide dog, or you, and it will never say “it’s safe to cross.”
Assists everything. Replaces nothing.
This is an assistive tool, not a mobility aid — it will never say “it’s safe to cross,” and it steps back whenever it’s unsure. It’s built on one principle: it assists your judgment — it never replaces your cane, your guide dog, or you. ChatLemur is built by a blind founder, and this app is designed at accessibility’s hardest setting — which is why accessibility isn’t different here. It’s better, for everyone.
Trust, earned — not claimed
Navigation software for a blind person has to earn trust before it gets used, not after. So the safety work isn’t a footnote here — it’s the headline, and every claim on this page is labelled honestly: proven, provisional, or not yet.
A Safety Envelope, not a promise
Steering cues ship disabled by default. A runtime Safety Envelope keeps them suppressed until on-device calibration passes — the code cannot steer anyone until that gate is earned.
2,300+ automated tests
The perception layer is built and locked by more than 2,300 automated tests that exercise the real wiring.
Adversarial safety reviews
The safety core is deliberately attacked by review, not just tested for the happy path — verified findings are fixed and locked in place by tests.
A readiness ledger, in public
Every capability is tracked as proven-in-code, provisional, or not-yet — the status badges on this page are that ledger, made public, not marketing copy.
Blind-led
Built and used daily by a blind founder — the person who will walk behind these safety rules first, before anyone else does.
Where each layer stands — and how it earns release
Navigation will be validated in stages with an Orientation & Mobility specialist before anyone walks with it unsupervised.
It sees and describes
BuiltScenes, text, objects, colours and faces — spoken to you, in real time, including continuous Live Scene narration and finger-targeted Point & Speak text reading. The perception engine is built and locked by more than 2,300 automated tests, running in our internal development builds — on-device validation is the next step.
In development: route-following assistance
In developmentSpoken route cues with careful awareness of the path right in front of you — steps, kerbs, obstacles. This layer is designed and in development to assist route-following after supervised validation; it will be validated in stages with an Orientation & Mobility specialist before anyone walks with it unsupervised.
Hands-free, on your face
Vision / futureA possible future interface: the same vision engine running on smart glasses — the camera as your eyes, hands-free. This is an exploratory direction, not a roadmap promise. The phone is primary today; no special hardware is required.
The road to release, in order
- Perception first. The seeing-and-describing layer is built and test-locked, in a development build.
- Navigation in development. Route cues and path awareness — steps, kerbs, obstacles — are in active development, not yet available.
- Shadow mode before anyone relies on it. The first field use is record-only: the app logs every cue it would have spoken while delivering none, so its judgment can be audited against reality before it ever guides a walk.
- Supervised, staged validation. The navigation layer will be validated in stages with an Orientation & Mobility specialist before anyone walks with it unsupervised.
- Only then, release. Nothing here is available to download yet — and we say so, everywhere it appears.
Assistive aid only. The live-navigation and edge-awareness parts are in development and will be validated with mobility specialists before any blind person walks with them. Your cane, your guide dog and your own judgment always come first.
Lenny is live on the web today.
While the Vision app earns its release the careful way, you can already talk to Lenny in your browser — free, no sign-up — and hear the same accessibility-first design working right now.